Racial differences in access to mortgage lending: comparison across major institutions

We use the 2019 Home Mortgage Disclosure Act (HMDA) dataset to measure racial bias in residential loan approval across the 20 largest mortgage granting institutions in the US. This data set reflects the changes in reporting requirements implemented from 2018 onwards, and include important variables that were missing from previous HMDA public samples. We observe large variation in unexplained race gaps across institutions, ranging 1.9–13.2% points for Black applicants, and 0.2–8.1% points for Hispanic applicants. The nationwide estimates are close to the middle of these ranges, and therefore mask the wide variation across institutions. We argue that comparison across major institutions reduces the impact of omitted-variable bias, which cast doubt on previous findings of “discrimination” based on HMDA public data. For such cross-institutions gaps to be driven by omitted variables, it would require unusual sorting of minority applicants across institutions. Our findings of possible discrimination are quite compelling due to (i) newly available variables, (ii) comparison across institutions, and (iii) the use of Oaxaca–Blinder–Fairlie decomposition, which reduces the bias due to the omission of variables correlated with race. We advocate for internal and external fair-lending reviews of institutions, starting with the ones at the top of our list in terms of unexplained race gaps and for publicizing the credit approval formulas used by institutions.